hand model based on yolov10 new
About
Summary
This algorithm is a deep learning-based hand detection system designed for clinical and surgical applications. It was developed using the YOLOv10 architecture, starting from pretrained weights and further fine-tuned on a curated dataset of annotated medical hand images. Training involved progressive optimization and data augmentation strategies to improve robustness and generalization. The algorithm outputs accurate localization of left and right hands, supporting downstream applications such as surgical workflow analysis, gesture recognition, and intraoperative assistance.
Mechanism
Target population: The algorithm targets surgical and clinical settings where accurate hand localization is required, particularly in operating rooms and interventional procedures.
Algorithm description: The model is built on YOLOv10, a state-of-the-art object detection framework optimized for speed and accuracy. It was initialized with pretrained weights and fine-tuned on a medical dataset with domain-specific augmentations. Optimization techniques such as AdamW, learning rate scheduling, weight decay, and regularization were applied to enhance model stability and reduce overfitting.
Inputs and Outputs:
Inputs: Static medical images or video frames containing clinical scenarios.
Outputs: Bounding boxes with class labels (“left hand” and “right hand”) and associated confidence scores. These predictions can be integrated into higher-level clinical applications, including real-time hand tracking, workflow monitoring, and decision-support systems.
Interfaces
This algorithm implements all of the following input-output combinations:
Validation and Performance
Challenge Performance
| Date | Challenge | Phase | Rank |
|---|---|---|---|
| Aug. 22, 2025 | t3challenge25 | Final Test Phase Task 2 | 1 |
Uses and Directions
This algorithm was developed for research purposes only.